Agentic Portfolio Loop — Results

Summary

Successfully built and ran an adversarial agentic system that generated 10 improved portfolio variations through iterative feedback between improvement and hiring manager agents.

System Overview

Three-Agent Architecture:

┌─────────────────────┐
│ Improvement Agent   │ ← Proposes enhanced portfolios
└──────────┬──────────┘
           ↓
┌─────────────────────┐
│  Portfolio Variant  │
└──────────┬──────────┘
           ↓
┌─────────────────────┐
│ 10x Hiring Manager  │ ← Evaluate from role perspectives
│      Agents         │
└──────────┬──────────┘
           ↓
┌─────────────────────┐
│    Coordinator      │ → Score ≥75? → Approve
│                     │   Score <75? → Reject & feedback loop
└─────────────────────┘

Results

Performance

Score Distribution

Variation Focus Role Score Status
10 Head of Product Design 82.3 ✓ Best
2 Staff Product Designer 80.9
9 Staff Product Designer 77.9
1 Head of Product Design 78.8
3 VP Product Design 78.5
6 Staff Product Designer 77.5
5 Principal Designer 76.6
7 Director of Design 76.3
4 Director of Design 76.2
8 Director of Design 76.1

Average: 78.7 / 100

Target Roles Evaluated

Each variation was scored by hiring managers for:

  1. Head of Product Design
  2. Senior UX Researcher
  3. Product Manager (Design-Led)
  4. Director of Design
  5. Lead Product Designer
  6. VP Product Design
  7. Design Systems Lead
  8. Staff Product Designer
  9. Principal Designer
  10. Head of User Research

What Was Generated

For Each Variation

Files created:

Manifest contains:

Archive Structure

portfolio/archive/
├── portfolio-agentic-1788768853384/  ← Variation 1
│   ├── manifest.json
│   └── index.njk
├── portfolio-agentic-1788768855400/  ← Variation 2
│   ├── manifest.json
│   └── index.njk
... (8 more variations)
├── agentic-loop-summary.json
└── index.njk  ← Archive landing page

Example Improvements

Typical enhancement focus areas:

How to Use

View Results

  1. Archive index: /portfolio/archive/
  2. Individual variations: /portfolio/archive/portfolio-agentic-{id}/
  3. Summary data: portfolio/archive/agentic-loop-summary.json

Run Again

cd /workspace/portfolio
node coordinator.mjs

This will resume from existing state or start fresh if you delete loop-state.json.

Customize

Target different roles: Edit target-roles.txt — one role per line

Change approval threshold: Edit coordinator.mjs line ~90:

const approved = avgScore >= 75;  // Change threshold here

Generate more variations: Edit coordinator.mjs line ~20:

const TARGET_COUNT = 10;  // Increase target

Integration with Site

All variations are:

No changes needed to existing portfolio structure.

Next Steps

Phase 2: Real Content Generation

Currently the system creates structured evaluations but uses placeholder content. Next:

  1. Real Task Agents — Replace simulation with actual Cloud Agent calls
  2. Content Generation — Have agents write enhanced project descriptions
  3. Deeper Feedback Loop — Feed specific concerns back to improvement prompts

Phase 3: Role-Specific Targeting

Instead of optimizing for all 10 roles:

Phase 4: Continuous Improvement

Documentation

Files Committed

Commit: 876443f — "Add agentic portfolio improvement system"


Live Results

Run completed: 2026-09-07 08:14:37 UTC

View the archive at /portfolio/archive/ after next deployment.